Krylov Directions Reveal Where Quantum Fisher Information Hides

Quantum Fisher information has quietly become one of the most versatile quantities in modern physics. It sets the ultimate precision limit of quantum metrology through the quantum Cramér–Rao bound, witnesses multipartite entanglement, signals quantum critical points, characterizes nonlocality, and probes the quantum geometry of state space. Yet for all its reach, the quantity has remained computationally stubborn: evaluating it requires solving a response problem in an operator space whose dimension grows quadratically with the Hilbert-space dimension. In interacting many-body systems, that quadratic blow-up rapidly becomes prohibitive, and the ultimate precision limit of quantum sensing has been inaccessible for anything but the smallest models.

A team of theoretical physicists at the Institute for Research in Fundamental Sciences in Tehran—M. Alishahiha, F. T. Tabesh, and M. J. Vasli—has now given this quantity a strikingly geometric makeover. Writing in The European Physical Journal C, they establish a direct correspondence between the Krylov hierarchy of quantum Fisher information approximations and a level-resolved decomposition of the symmetric logarithmic derivative, the operator whose state-dependent norm defines the Fisher information itself. The result is a genuine probability distribution—the Krylov distribution—that tells you exactly where, in operator space, the metrological sensitivity of a quantum state resides.

The technical heart of the problem is deceptively simple. For a differentiable family of states, the symmetric logarithmic derivative is defined implicitly by the equation in which the anticommutator superoperator acting on the derivative gives the tangent operator. For unitary parameter encoding, the tangent is the commutator of the state with the Hamiltonian, and the quantum Fisher information is the state-dependent norm of the derivative. Evaluating the Fisher information is therefore equivalent to solving a linear-response problem in Liouville space, where the superoperator plays the role of the matrix and the tangent plays the role of the source. The difficulty is that this operator space is the square of the physical Hilbert space, so direct construction or inversion of the superoperator becomes prohibitive in interacting many-body systems.

Krylov-subspace methods offer a way out. Instead of solving the response problem globally, the solution is approximated in a nested sequence of subspaces generated by repeated application of the superoperator to the tangent, with the Lanczos algorithm constructing an orthonormal basis and a tridiagonal representation of the operator. The Tehran team’s central identity shows that the quantum Fisher information resolved at each Krylov level is exactly the incremental gain of the Lanczos hierarchy: the squared coefficients of the derivative in a filtration-adapted state-orthonormal Krylov basis equal the differences between successive approximations. A single Lanczos run in the Hilbert–Schmidt geometry therefore yields the complete level-resolved information distribution, giving the Krylov distribution a direct information-theoretic interpretation.

The payoff is that convergence acquires a physical meaning. The truncation error of the hierarchy is not merely a numerical residual; it is the unresolved probability weight residing in higher Krylov directions. The first moment of the distribution defines an intrinsic information depth—a measure of how many Krylov directions are required, on average, to resolve the metrological response. A small depth indicates that the Fisher information is concentrated in response directions close to the tangent operator, while a large depth signals that a substantial fraction of the sensitivity is distributed over more deeply generated directions. The team also derives a spectrum-independent structural guarantee: once the Krylov dimension is large compared with the information depth, only a small fraction of the Fisher information can remain unresolved.

The framework’s most consequential new result concerns the singular regime. When the active spectrum of the superoperator is gapped—contained strictly away from zero—the inverse response is analytic on its support and the hierarchy converges exponentially, with a rate controlled by the condition number of the tangent-active spectrum. But when the spectral support reaches the origin, the picture changes qualitatively. For an exactly solvable Jacobi family of spectral measures, the team derives an exact finite-order convergence law: the relative error equals the square of a ratio of gamma functions, which decays algebraically as a power of the Krylov order, with an exponent set by the edge parameter. The marginal case at the boundary of the family marks the transition where the Fisher information becomes logarithmically divergent, separating finite from divergent metrological susceptibility.

The hard-edge analysis also delivers a subtle surprise: finite Fisher information does not necessarily imply a finite information depth. For edge parameters between one and three-halves, the total Fisher information remains finite, but the mean Krylov depth diverges—the normalized information profile has a tail so long that its mean level is infinite. For every fixed tolerance, however, the error still falls below that tolerance at a finite Krylov order, so divergent mean depth is not failure of finite-tolerance convergence. The team is careful to frame the Jacobi result as an analytically controlled benchmark rather than a universal theorem: more general spectral densities require additional regularity assumptions before analogous convergence laws can be established.

To demonstrate the framework in action, the authors implemented a matrix-free numerical calculation for a mixed-field Ising system with twelve spins, sampling full-rank states from a random Ginibre ensemble. Working in the eigenbasis of the density matrix, the Fisher information kernel becomes a diagonal spectral problem, allowing the Lanczos recursion to be performed without ever constructing the Liouville-space superoperator. The numbers are remarkable. The compressed spectral space comprises 8,386,560 active pairs, yet the quantum Fisher information reaches a relative resolution of ten to the minus fifteen after only about 4,500 Krylov directions—roughly 0.054 percent of the maximum pair-space dimension.

The ensemble-averaged results reveal two distinct features of the same distribution. The first three Krylov directions already carry about 97.6 percent of the Fisher information weight, and the resolved information depth remains well below unity, reflecting the overwhelming concentration of metrological weight at very low Krylov order. At the same time, the hierarchy possesses a much longer but extremely weak tail: the calculation crosses the operational threshold only at about 4,500 directions. A distribution can have a very small mean depth while still possessing a long, numerically tiny tail, and the large separation between the two scales is therefore not contradictory but a signature of the strong intrinsic compressibility of the metrological response in Krylov space.

The authors emphasize that the finite-size numerics realize the gapped regime—every full-rank state has a strictly positive spectral minimum—while the hard-edge algebraic regime would require an accumulation of eigenvalues toward zero, as may occur near quantum critical points, in thermodynamic limits, or at low temperatures. The relevant edge exponent must be determined from the tangent-weighted pair spectrum, not from the density of eigenvalues of the state alone, and the relevant gap is a gap of the superoperator, not the excitation gap of the Hamiltonian. For an exactly pure state with unitary encoding, the hierarchy saturates after a single step, irrespective of whether the generator is gapless. The operator-geometric perspective developed here is not restricted to quantum Fisher information and may extend to other quantum information quantities defined through operator inverse problems, potentially offering a unified geometric language for where quantum information resides and how it is progressively resolved.

Subject of Research: Krylov distribution and spectral convergence of quantum fisher information

Article Title: Krylov distribution and spectral convergence of quantum fisher information

Article References: Alishahiha, M., Tabesh, F. T., & Vasli, M. J. (2026). Krylov distribution and spectral convergence of quantum fisher information. The European Physical Journal C, 86(10), Article 1176. https://doi.org/10.1140/epjc/s10052-026-16434-0

Image Credits: AI Generated

DOI: 10.1140/epjc/s10052-026-16434-0

Keywords: quantum Fisher information, Krylov complexity, quantum metrology, Lanczos algorithm, quantum information, many-body physics, spectral convergence, operator growth, Krylov, distribution, spectral, convergence